https://www.linkedin.com/pulse/types-relationships-object-oriented-programming-oop-sarah-el-dawody/

Composition:是一种“part-of”关系。

例如，“发动机是汽车的一部分”，“心脏是身体的一部分”。

关联:是“has-a”类型的关系

例如，假设我们有两个类，那么这两个类被称为“has-a”关系，如果这两个实体为了某些工作而共享彼此的对象，同时它们可以不依赖彼此而存在，或者它们都有自己的生命期。

上面的示例显示了一种关联关系，因为Employee类和Manager类都使用彼此的对象，并且都有各自独立的生命周期。

聚合:是基于“has-a”关系，是一种特殊的关联形式

例如，“学生”和“地址”。每个学生必须有一个地址，所以学生类和地址类之间的关系将是“Has-A”类型的关系，反之亦然。

这些答案的问题在于，它们只说了一半:它们解释了聚合和组合是关联的形式，但没有说一个关联是否可能不是这两种形式。

基于对SO和一些UML文档的一些简要阅读，我收集到类关联有4种主要的具体形式:

A是由A组成的;没有A, B就不存在，就像家里的房间一样 聚合:A有A B;B可以没有A而存在，就像教室里的学生一样 依赖:A使用B;A和B之间没有生命周期依赖关系，比如方法调用参数、返回值或方法调用期间创建的临时对象 泛化:A是A

当两个实体之间的关系不是其中之一时，它可以被称为一般意义上的“关联”，并以其他方式进一步描述(注意，原型等)。

我的猜测是，“通用关联”主要用于两种情况:

when the specifics of a relationship are still being worked out; such relationship in a diagram should be converted as soon as possible to what it actually is/will be (one of the other 4). when a relationship doesn't match any of those 4 predetermined by UML; the "generic" association still gives you a way of representing a relationship that is "not one of the other ones", so that you aren't stuck using an incorrect relationship with a note "this is not actually aggregation, it's just that UML doesn't have any other symbol we could use"

    Simple rules:
    A "owns" B = Composition : B has no meaning or purpose in the system 
    without A
    A "uses" B = Aggregation : B exists independently (conceptually) from A
    A "belongs/Have" B= Association; And B exists just have a relation
    Example 1:

    A Company is an aggregation of Employees.
    A Company is a composition of Accounts. When a Company ceases to do 
    business its Accounts cease to exist but its People continue to exist. 
    Employees have association relationship with each other.

    Example 2: (very simplified)
    A Text Editor owns a Buffer (composition). A Text Editor uses a File 
    (aggregation). When the Text Editor is closed,
    the Buffer is destroyed but the File itself is not destroyed.

摘自:Remo H. Jansen的《Beginning React: Learning TypeScript 2》。x -第二版”:

We call association those relationships whose objects have an independent life cycle where there is no ownership of the objects. Let's take a look at an example of a teacher and a student. Multiple students can be associated with a single teacher, and a single student can be associated with multiple teachers, but both have independent life cycles (both can create and delete independently). So, when a teacher leaves the school, we don't need to delete any students, and when a student leaves the school, we don't need to delete any teachers.

我们称这些关系为聚合，其对象具有独立的生命周期，但存在所有权，并且子对象不能属于另一个父对象。让我们以手机和电池为例。单个电池可以属于一个手机，但如果手机停止工作，我们从数据库中删除它，手机电池将不会被删除，因为它可能仍在工作。因此，在聚合中，虽然有所有权，但对象有它们的生命周期

我们使用术语组合来指代对象没有独立生命周期的关系，如果父对象被删除，所有子对象也将被删除。让我们以问题和答案之间的关系为例。一个问题可以有多个答案，答案不能属于多个问题。如果我们删除问题，答案将自动删除。

在面向对象编程中，类是相互关联的。这意味着它们的实例相互调用方法。因此，如果一个类的实例调用另一个类的方法，它们是相关的，通常我们用ASSOCIATION来建模这种关系。 例如，在下面的代码片段中，Customer类与Order类相关联。她/他取消了订单。

class Customer {
        private Order[] orders;
        public boolean removeCart() {
                for (int i = 0 ; i < orders.length ; i++) {
                        orders[i].cancel();
                }
        }
}

AGGREGATION意味着一个类拥有另一个类的一些实例。它只不过是联想，马丁·福勒建议不要使用它。因为当一个类与另一个类相关联时，它有一个对该类的引用来调用该类上的方法。

但是COMPOSITION是关联的一个有意义的子集。这意味着一个类是由其他一些类组成的。例如，我们有一个学生类，由其他一些类组成，如ReportCard。我们知道成绩单是非常依赖于学生的，如果我们从系统中删除了学生，他们的成绩单也应该被删除。

令人惊讶的是，关于关联、聚合和组合这三个关系概念之间的区别存在如此多的混淆。

请注意，术语聚合和组合已经在c++社区中使用，可能在它们被定义为UML类图中关联的特殊情况之前已经有一段时间了。

主要的问题是广泛的和持续的误解(甚至在专家软件开发人员中)，组合的概念意味着整体和它的部分之间的生命周期依赖关系，以至于部分不能没有整体而存在，忽略了这样一个事实，即也存在与不可共享部分的部分-整体-关联的情况，其中部分可以从整体中分离出来，并且在整体被破坏后仍然存在。

在我看来，这种困惑有两个根源:

In the C++ community, the term "aggregation" was used in the sense of a class defining an attribute for referencing objects of another independent class (see, e.g., [1]), which is the sense of association in UML Class Diagrams. The term "composition" was used for classes that define component objects for their objects, such that on destruction of the composite object, these component objects are being destroyed as well. In UML Class Diagrams, both "aggregation" and "composition" have been defined as special cases of associations representing part-whole relationships (which have been discussed in philosophy for a long time). In their definitions, the distinction between an "aggregation" and a "composition" is based on the fact if it allows sharing a part between two or more wholes. They define "compositions" as having non-shareable (exclusive) parts, while "aggregations" may share their parts. In addition they say something like the following: very often, but not in all cases, compositions come with a life-cycle dependency between the whole and its parts such that the parts cannot exist without the whole.

因此，尽管UML将术语“聚合”和“组合”放在了正确的上下文中(部分-整体关系)，但是他们并没有设法以一种清晰和明确的方式来定义它们，从而捕捉开发人员的直觉。然而，这并不奇怪，因为这些关系可以有很多不同的属性(和实现的细微差别)，开发人员对如何实现它们没有一致的意见。

请参见我对2009年4月SO问题的扩展回答。

c++社区中定义OOP对象之间“组合”的属性(这个信念仍然被广泛持有):两个相关对象(组合及其组件)之间的运行时生命周期依赖关系并不是“组合”的真正特征，因为在其他类型的关联中，我们也可以由于引用完整性而具有这种依赖关系。

例如，在一个SO回答中提出了以下“composition”的代码模式:

final class Car {    
  private final Engine engine;

  Car(EngineSpecs specs) {
    engine = new Engine(specs);
  }

  void move() {
    engine.work();
  }
}

被调查者声称这是“合成”的特征，没有其他类可以引用/知道这个组件。然而，并非所有可能的“组合”情况都是如此。特别是，在汽车引擎的情况下，汽车的制造商(可能是在另一个类的帮助下实现的)可能必须引用引擎，以便在出现问题时能够联系汽车的所有者。

[1] http://www.learncpp.com/cpp-tutorial/103-aggregation/

附录-关于StackOverflow上的组合与聚合的反复询问问题的不完整列表

[Apr 2009] Aggregation versus Composition [closed as primarily opinion-based by] [Apr 2009] What is the difference between Composition and Association relationship? [May 2009] Difference between association, aggregation and composition [May 2009] What is the difference between composition and aggregation? [duplicate] [Oct 2009] What is the difference between aggregation, composition and dependency? [marked as duplicate] [Nov 2010] Association vs. Aggregation [marked as duplicate] [Aug 2012] Implementation difference between Aggregation and Composition in Java [Feb 2015] UML - association or aggregation (simple code snippets)